It’s actually $16 million per drone, still a whopping amount of money.
The analogy to early World War II torpedoes is perhaps misapplied. The programmatic problem with World War II torpedoes was inadequate testing. Once combat had revealed the problems that testing should have, the shipyard at Pearl Harbor performed their own tests and was able to identify the problems and fix them relatively cheaply and easily.
The problem with these drones may be resolved by tweeking a few software parameters, or the concept may be flawed and the whole program is a waste of taxpayer money. I suspect the truth lies somewhere in between.